Learning Convective Cloud Regimes over the Asian Monsoon Area

One of the key features during Asia summer monsoon is the occurrence of intense rainfall and mesoscale convective systems, which is tightly coupled with seasonal transition of large-scale environment and can exhibits multi-scale variability. . Previous observational studies based on satellite and ground radar have identified that convective clouds with different structures are triggered by particular environmental conditions over specific regions. In this work, we explore the application of machine learning (ML) to objectively classify convective cloud regimes over Asian monsoon areas observed in CloudSat Vertical Cloud Mask over 2006-2015 and investigate their spatial-temporal variation of different regimes. Using the 6-way connected method and the vertical continuity criteria, convective cloud are defined as the connected cloudy pixels in CloudSat track, and the object-based cloud properties can be recorded. These convective clouds are then objectively classified by five cloud features (horizontal scale, 2d vertical cloud size, cloud top, and convective strength: radar reflectivity 0dBz and 10dBz reach height) using the agglomerative clustering algorithm, which is an unsupervised, hierarchical cluster analysis. The clustering results yield five types of convective cloud regimes with distinct cloud properties. The first type (coastal-medium) occurs more frequently over coastal area, with averaged horizontal scale over 500 km and medium convective strength (0 dBz echo height > 10 km in 80% of the cases). The second type (coastal-intense) also appears mainly over coastal area but with significant seasonal variability that coincides tightly with the sharp onset of the Asian summer monsoon. Most of the coastal-intense type are organized systems which have very large horizontal scale (averaged over 1000 km) and high convective strength (10 dBz echo height > 10 km). The other three types of convective cloud regimes have smaller horizontal scale (averaged < 200 km) and share similar spatial distribution over tropical open ocean and inland areas, while the main difference between these types is their convective strength. In the future the association of these convective cloud regimes with large-scale climate variability and local environmental conditions over their occurrence hotspots will the explored using high-resolution reanalysis data.